US20100179803A1 - Hybrid machine translation - Google Patents

Hybrid machine translation Download PDF

Info

Publication number
US20100179803A1
US20100179803A1 US12/606,110 US60611009A US2010179803A1 US 20100179803 A1 US20100179803 A1 US 20100179803A1 US 60611009 A US60611009 A US 60611009A US 2010179803 A1 US2010179803 A1 US 2010179803A1
Authority
US
United States
Prior art keywords
statistical
based
translation
rule based
machine translation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
US12/606,110
Other versions
US9798720B2 (en
Inventor
Hassan SAWAF
Mohammad Shihadah
Mudar Yaghi
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
eBay Inc
Original Assignee
AppTek
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to US19305808P priority Critical
Application filed by AppTek filed Critical AppTek
Priority to US12/606,110 priority patent/US9798720B2/en
Assigned to AppTek reassignment AppTek ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: SAWAF, HASSAN, SHIHADAH, MOHAMMAD, YAGHI, MUDAR
Publication of US20100179803A1 publication Critical patent/US20100179803A1/en
Assigned to SCIENCE APPLICATIONS INTERNATIONAL CORPORATION reassignment SCIENCE APPLICATIONS INTERNATIONAL CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: APPLICATIONS TECHNOLOGY, INC.
Assigned to LEIDOS, INC. reassignment LEIDOS, INC. CHANGE OF NAME (SEE DOCUMENT FOR DETAILS). Assignors: SCIENCE APPLICATIONS INTERNATIONAL CORPORATION
Assigned to APPLICATIONS TECHNOLOGY (APPTEK), LLC reassignment APPLICATIONS TECHNOLOGY (APPTEK), LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: LEIDOS, INC.
Assigned to APPLICATIONS TECHNOLOGY (APPTEK), LLC reassignment APPLICATIONS TECHNOLOGY (APPTEK), LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: SAWAF, HASSAN, SHIHADAH, MOHAMMAD, YAGHI, MUDAR
Assigned to EBAY INC. reassignment EBAY INC. ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: APPLICATIONS TECHNOLOGY (APPTEK), LLC
Assigned to APPLICATIONS TECHNOLOGY (APPTEK), LLC reassignment APPLICATIONS TECHNOLOGY (APPTEK), LLC ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: SAWAF, HASSAN, SHIHADAH, MOHAMMAD, YAGHI, MUDAR
Publication of US9798720B2 publication Critical patent/US9798720B2/en
Application granted granted Critical
Application status is Active legal-status Critical
Adjusted expiration legal-status Critical

Links

Images

Classifications

    • G06F40/40

Abstract

A system and method for hybrid machine translation approach is based on a statistical transfer approach using statistical and linguistic features. The system and method may be used to translate from one language into another. The system may include at least one database, a rule based translation module, a statistical translation module and a hybrid machine translation engine. The database(s) store source and target text and rule based language models and statistical language models. The rule based translation module translates source text based on the rule based language models. The statistical translation module translates source text based on the statistical language models. A hybrid machine translation engine, having a maximum entropy algorithm, is coupled to the rule based translation module and the statistical translation module and is capable of translating source text into target text based on the rule based and statistical language models.

Description

    BACKGROUND OF THE INVENTION
  • An alternative to human translation of expression from one language into another, machine translation has been applied to increase translation throughput. In this regard, computing power has advanced exponentially over the past three decades to the point where intelligence analysts and linguists now can use powerful tools to assist them in processing large volumes of disparate data from multiple sources in real time. Machine translation (MT) is a software-based technology that can aid linguists and analysts in processing the volumes of incoming information whether from print, electronic, audio and video sources. There are two distinct methods for machine translation, rule-based and statistical. Each one has its advantages, but no product exists that combines the best of both worlds for a hybrid machine translation solution.
  • In this application, we describe a hybrid machine translation system, based on a statistical transfer approach using statistical and linguistic features and highlight the system's capabilities on applications of o machine translation in different tasks:
  • (a) Translation of one language into another for very large vocabulary broadcast, newswire and web texts. Their input is either captured from the Internet or is recorded form a satellite feed and recognized using a speech recognition system; and
  • (b) Translation of one Language into another for medium to large vocabulary speech-to-speech translation. The input is recorded through a telephone channel and recognized using an automatic speech recognition system.
  • The recognized utterances and the text captured from the internet are normalized, using statistical machine translation that is based on finite state automata. The output of this interlingua is then translated by a hybrid machine translation system, combining statistical and rule-based features. This application introduces also a hybrid interlingua approach that gives better results for dialect speech input compared to a direct machine translation system based on a statistical approach and a direct machine translation based on a rule-based approach.
  • Applying Machine Translation
  • The current process for handling information is largely a manual one. Intelligence operations are highly reliant on the skills of the people performing foreign language translations and on those analyzing and interpreting the data while the volume of data grows exponentially and the pool of qualified people continues to shrink. Machine translation tools exist to assist linguists and analysts in doing their job.
  • What is Machine Translation?
  • Machine translation (MT) involves the use of computer software to translate one natural human language into another. MT takes into account the grammatical structure of each language, and uses contextual rules to select among multiple meanings, in order to transfer sentences from the source language (to be translated) into the target language (translated).
  • MT refers to the process of translating a variety of media (speech, text, audio/video, web pages, etc.) from one language to another using computers and software. MT is designed to support and assist intelligence analysts and linguists with their human translation tasks.
  • Translation, in its simplest definition, involves: decoding the meaning of the source text; and re-encoding this meaning in the target language. A translator decodes the meaning of the source text in its entirety. This means that the translator must interpret and analyze all the features of the text by applying in-depth knowledge of the grammar, semantics, syntax, idioms, and the like of the source language, as well as the culture of its speakers. At the same time, the translator needs an equivalent in-depth knowledge to re-encode the meaning in the target language.
  • Foreign language translation can be difficult even for a skilled linguist. Performing the same translations using machine translation increases the accuracy and speed of translating text and identifying key points of interest. The question is: How do you program a computer to “understand” a text just as a person does, and also to “create” a new text in the target language that “sounds” as if it has been written by a person? Machine translation software is designed to address this problem through two main approaches: a rules-based approach, and a statistical approach.
  • Rule-Based Machine Translation
  • A rules-based approach is a method based on linguistic rules meaning that words will be translated in a linguistic way, that is, the most suitable (orally speaking) words of the target language will replace the ones in the source language.
  • Generally, rule-based methods parse a text, usually creating an intermediary, symbolic representation, from which the text in the target language is generated. These methods require extensive lexicons with morphological, syntactic, and semantic information, and large sets of rules.
  • With sufficient data, MT programs often work well enough for a native speaker of one language to get the approximate meaning of what is written by the other native speaker. The difficulty is getting enough data of the right kind to support the particular method.
  • Rule-based translation approaches have the advantage of a high abstraction level, and allow understandable translations for a high coverage, i.e., the “informativeness” (the accurate translation of information) of the translation is higher for a higher coverage of domains and types of texts.
  • A prime motivation for creating a hybrid machine translation system is to take advantage of the strengths of both rule-based and statistical approaches, while mitigating their weaknesses. Thus, for example, a rule that covers a rare word combination or construction should take precedence over statistics that were derived from spares data (and therefore is not very reliable). Additionally, rules covering long-distance dependencies and embedded structures should be weighted favorably, since these constructions are more difficult to process in statistical machine translation.
  • Statistical Machine Translation
  • Statistical machine translation tries to generate translations using statistical methods based on a large body of bilingual text. Such an example is the Canadian Hansard corpus, the English-French record of the Canadian parliament. Ambiguity of some words can change the meaning and subsequent translation. Today, both “shallow” and “deep” approaches are used to overcome this problem. Shallow approaches assume no knowledge of the text; they simply apply statistical methods to the words surrounding the ambiguous word. Deep approaches presume a comprehensive knowledge of the word. Thus, a statistical approach should take precedence in situations where large numbers of relevant dependencies are available, novel input is encountered or high-frequency word combinations occur.
  • Today, no single system provides a “fully-automatic, high-quality machine translation.” Rule based machine translation applications have come the closest so far, however, there are some advantages of statistical machine translation that are not fully realized in pure rule based machine translation systems.
  • Accordingly, there remains a need for an optimal machine translation system. There is a further need for systems to perform translation using the best available models for translation of given source and target languages.
  • SUMMARY OF THE INVENTION
  • According to the present invention, a system and method for hybrid machine translation approach is based on a statistical transfer approach using statistical and linguistic features. The system and method may be used to translate from one language into another, for example: (i) translation of Modern Standard and “Slightly Dialectal” Arabic into English, for very large vocabulary Broadcast, Newswire and Web texts; and (ii) Iraqi and Modern Standard Arabic into English, for medium to large vocabulary speech-to-speech translation. The systems and methods use a workflow that incorporates technology to facilitate the processing of collected materials and, therefore, allow linguists and analysts to make more accurate decisions faster about what information is a potentially valuable asset to analyze and provide more accurate language translations of text and recognized speech inputs.
  • According to one embodiment of the invention, a system for automatic translation comprises at least one database, a rule based translation module, a statistical translation module and a hybrid machine translation engine. The database(s) store source and target text and rule based language models and statistical language models. The a rule based translation module translates source text based on the rule based language models. The statistical translation module translates source text based on the statistical language models. In addition, a hybrid machine translation engine, having a maximum entropy algorithm, is coupled to the rule based translation module and the statistical translation module and is capable of translating source text into target text based on the rule based and statistical language models.
  • The maximum entropy algorithm of the hybrid machine translation engine may be configured to assign probabilities to different statistical language models and different rule based models. In addition, the hybrid machine translation engine may be configured to perform translation of source text into target text based on available rule based and/or statistical language models.
  • According to additional embodiments of the invention, the database may further store directed acyclic graphs (DAGS) used by both the rule based and statistical translation systems and the rule based translation module may include semantic rules that are used to annotate the statistical DAGS. The rule based translation module may include rule based syntactic, semantic, and grammatical language models and a transfer system model. By contrast, the statistical translation module may include a statistical lexicon model, a statistical phrase model and a statistical language model. At least one of the syntactic, semantic and grammatical based models may be used to eliminate paths from a DAGS used by the hybrid machine translation engine.
  • According to another embodiment of the invention, a system for automatic translation of source text into target text, may comprise at least one database for storing source and target text and directed acyclic graphs (DAGS) and semantic rules. It may further include a statistical translation system that generates DAGS based on source text and a hybrid machine translation engine that annotates the DAGS based on semantic rules and translates source text into target text based on at least the DAGS. The hybrid machine translation engine may use a maximum entropy algorithm to perform the translation.
  • According to still another embodiment of the invention, a method for machine translation comprises receiving and storing source text, generating directed acyclic graphs (DAGS) based on statistical models, annotating the DAGS based on semantic rules and translating the source text into target text based at least in part on the annotated DAGS. The translating may be performed based on a maximum entropy algorithm that further receives weighted inputs from statistical and rule based translation system models and performs the translating based on the weighted inputs.
  • BRIEF DESCRIPTION OF THE FIGURES
  • The above described features and advantages of the present invention will be more fully appreciated with reference to the detailed description and accompanying drawing figures.
  • FIG. 1 depicts a hybrid machine translation system according to one embodiment of the present invention.
  • FIGS. 2A-2C depict flow charts of a hybrid machine translation and training process according to an embodiment of the present invention.
  • FIG. 3 depicts a word graph.
  • FIG. 4 depicts a method of annotating a DAG and performing a hybrid name translation based on the annotated DAG according to an embodiment of the invention.
  • FIG. 5 depicts a method of performing hybrid machine translation on a text source and a source of text from speech recognition output according to an embodiment of the invention.
  • FIGS. 6A, 6B and 6C depict another view of a hybrid machine translation system according to an additional embodiment of the invention.
  • DETAILED DESCRIPTION Approaches to Machine Translation
  • Machine Translation can be done in different manners, which can be roughly put in two categories: corpus-based methods and rule-based methods. Both approaches have advantages and disadvantages:
  • Corpus-based methods—namely example-based and statistical methods—are very fast to develop and they are very consistent, as they use examples to learn from, the disadvantage is that the approach needs a certain amount of data to extract the necessary information from the examples and the translation is only then of high quality, as long as the sentences are of the same domain and similar type as the sentences in the training data, i.e. they show high grades of “fluency”.
  • Rule-based translation approaches have the advantage, that they have a high abstraction level, and allow understandable translations for a high coverage, i.e. the “informativeness” of the translation is higher for a higher coverage of domains and types of texts.
  • To combine the qualities of both approaches, a hybrid machine translation approach as described herein may be used. The use of parsing features in machine translation showed a quality boost in the past, as can be read in (Sawaf et al. 2000). But beyond the use of syntax-based language models, we want to use more linguistic information in the machine translation. Our approach is based on the mathematical basis of corpus-based approaches, namely the statistical approach described in (Ney et al. 2000).
  • FIG. 1 shows the use of a hybrid machine translation engine 100 which makes use of a maximum entropy algorithm 120 implemented by a hybrid machine translation engine 100 and inputs from statistical and rule based models. The statistical inputs include, for example, a statistical lexicon model 130, a statistical phrase model 140 and a statistical language model 150. The rule based inputs include, for example, a rule based syntactic model 160, a rule based syntactic model 170, a rule based grammatical model 180 and a rule based transfer system 190. The output of each of the models may be weighted and the maximum entropy algorithm may take into account these weights when selecting the optimum translation.
  • FIG. 2A depicts a translation process. Referring to FIG. 2A, in step 200, a source sentence enters the translation process and generally a computer system for performing the translation process. The sentence may be a single text input sentence, a single transcription utterance, a n-best text input in list form, a n-best text input in DAG form, a n-best transcription in list form, a n-best transcription in DAG form or any other source of text. Text transcription may be input from a speech recognition process, a character recognition process or other process of transcribing text. The text may be stored in memory for processing.
  • In step 202, the computer preprocesses the text. The preprocessing process is shown in FIG. 2B. The out put of the preprocessing process is may be sent for rule based machine translation processing in step 204 and statistical machine translation processing in step 206. In step 208, a hybrid machine translation processes receives inputs from the rule-based machine translation 204 and statistical machine translation processing 206 and outputs translation information to step 210. In step 210, the computer performs postprocessing and outputs in step 212 the output of the translation process. The output may be one or more translated sentences, including, for example, one best hypothesis, n-best hypotheses in list form or n-best hypotheses in DAG form.
  • FIG. 2B shows a preprocessing process according to an embodiment of the invention. Referring to FIG. 2B, in step 222 a dialect normalization and noise elimination step receives a source sentence, for example from the translation process shown in FIG. 2A. In step 224, a computer performs named entity recognition on the source text. Then in step 226, information from steps 220-224 are stored in memory in step 226 for translation and in step 228, translation of named entities is performed with resulting information being saved in memory.
  • In step 230, a morphological analyzer analyzes the information stored in memory and in step 232 reordering is done on the source sentence. In step 234, the preprocessed source sentence is available for the translation processes shown in FIG. 2A and other translation processes shown and described herein.
  • FIG. 2C depicts a translation training process according to an embodiment of the present invention. Referring to FIG. 2C, in step 240 source reference sentences are input to the training process. In step 242, a name finder is applied to the source, followed by a morphological analyzer in step 244. In step 246 a LFG parser and transfer process is applied, the output of which is sent to the AT extraction process 268. In step 248, a lexical lookup is used to utilize rule-based and lexical features of rule based machine translation. In step 252, target reference sentences are input that correspond to the source reference sentences. A name finder for the target is applied in step 256 and then a morphological analyzer processes the target text. In step 258, the text from the morphological analyzer is input to a LM training process, which further outputs data to a statistical language models database 264.
  • In step 250, a SMT training process, for example (GIZA++) receives inputs from processes 244, 248 and 256 and is used to generate a bilingual lexicon database 260 and an alignments database 262. In step 268 an alignment template extraction step is performed an alignment templates database is created in step 266. In this manner, various databases may be created through a training process that involves inputting source and target reference text and analyzing the source and reference text statistically and according to rule based translation to populate databases that are used by a hybrid matching translation database to perform machine translation.
  • Following is a more detailed discussion of certain terms and steps that appear in FIGS. 2A-2C and their application to hybrid machine translation.
  • Hybrid Machine Translation
  • The approach of the hybrid machine translation system according to one embodiment of the invention is based on the mathematical definition of machine translation—the implementation is based on cascaded finite-state automaton (FSA) as are used for Speech Recognition and Machine Translation and other human language technology applications. We refer to the approach as “stacked FSA MT approach”. The use of probabilistic FSAs for hybrid machine translation allows us to easily combine syntactical and other linguistically motivated features with statistical machine translation features and algorithms.
  • Dialect Normalization and Noise Elimination
  • To translate documents from the Internet, e.g. “Blogs” or emails, the need for noise reduction by normalization is of very high importance. The process is basically a monotone translation, allowing also phrases to be translated, so that special email jargon can be translated to standard text, as well as misspellings can be corrected. For this, we can use a standard statistical approach, and the use of a strong background lexicon to achieve good results in the spellchecking. The same process, with the same techniques can be used to do dialect normalization. The Arabic dialects are translated into Modern Standard Arabic. Currently, we distinguish 16 different main Arabic dialects: Modern Standard Arabic and colloquial Arabic dialects from the following regions: Lebanon, North Syria, Damascus, Palestine, Jordan, North Iraq, Baghdad, South Iraq, Gulf, Saudi-Arabia, South Arabic Peninsula, Egypt, Libya, Morocco, Tunisia. There is no limit to the number of dialects that may be used. Transferring the words from the dialects to Modern Standard Arabic can be seen as a Interlingua approach for dialect normalization.
  • Named Entity Recognition/Translation
  • To increase quality of the MT output, words which describe names, locations, events, dates, etc. (so-called “Named Entities”) should be handled differently than other words. Many times, Named Entities can also be categorized as adjectives or regular nouns, like
    Figure US20100179803A1-20100715-P00001
    , which is a male's first name “Kareem” (potentially also a last name) or an adjective “generous”, and of course, the translation of the word depends heavily on this categorization. Depending on the context, it can be determined, whether it is a Named Entity or not.
  • For this categorization, the following approach is used: In general, the decision is a search problem, which according to Bender et all (2003) can be described as:
  • Pr ? == exp ( m = 1 M ? ) ? exp ( m = 1 M ? ) ; ? indicates text missing or illegible when filed ( 1.2 )
  • () deleting so-called feature-functions, which are either linguistic, lexical, morphological or statistically learned phenomena. An example of such feature is the fact that a word like “the doctor” (in Arabic
    Figure US20100179803A1-20100715-P00002
    indicates that the following word is most probably a name, if it is not a verb. An overview of some of the features can be seen in the following table:
  • TABLE 1 Some feature types as described in Bender et al. (2003) FEATURE TYPE DESCRIPTION Lexical words are in a fixed lexicon Word digits and numbers: date and time Transition distribution depending on predecessor words Compound distribution depending whether the word is part of a compound or not Signal Word signal words from a signal word lexicon could trigger. if in a specific proximity
  • The translation of a Named Entity may be dealt with differently than other words. Named Entities have three different ways of translation:
  • Translate the Named Entity by translating the linguistic (sub-)entities: e.g.
    Figure US20100179803A1-20100715-P00003
    Figure US20100179803A1-20100715-P00004
    Figure US20100179803A1-20100715-P00005
    into “Ministry of Foreign Affairs”;
  • Translate the Named Entity by using a previously seen transcription into the target language alphabet. Previously seen means seen in bilingual training data. E.g.
    Figure US20100179803A1-20100715-P00006
    into “Ahmad”;
  • Translate the Named Entity by using an automatic transcription approach, using a phonetic mapping of the source language graphemes into source language phonemes, then mapping the source language phonemes into target language phonemes, then finally converting these target language phonemes into the target language graphemes, which are the sub-units of the word (for Arabic, similar to many European languages, the graphemes are mainly the “letters” of the alphabet).
  • Morphological Analysis
  • In statistical analysis, the number of occurrences of words and the number of unique words in different environments is taken into account for the statistical models (to prevent the “sparse data problem”, meaning that phenomena—also called “events”—occur only sparsely and the statistics gathered around it is not sufficient to have a reliable modeling) as well as speed (smaller translation lexicon means fewer possible translations and fewer transactions to reach a translation). So, to deal with this class of problems, morphological analysis can be considered.
  • For languages like English, morphological analysis has only a minor role for decreasing the number of events for the modeling for syntactical analysis and/or generation. Other languages like French and Spanish have a higher morphological complexity, and the use of morphological analysis decreases the vocabulary size by about 10-20%. For a language like German, where the morphology analysis helps deal with the well-known compound word problem, the vocabulary size can go down by 15-35%.
  • For languages like Arabic, morphology plays a very big role. For tasks like the GALE task, with a original vocabulary size of about 400K words, the morphology can decrease the value by up to 70%, down to 120K morphemes. Table 1 shows examples of morphological analyses of different words.
  • TABLE 1 Morphology examples for Arabic words Full-form Word Morphemes Meaning
    Figure US20100179803A1-20100715-P00007
    Figure US20100179803A1-20100715-P00008
    Figure US20100179803A1-20100715-P00009
    Figure US20100179803A1-20100715-P00010
    (they) will move her
    Figure US20100179803A1-20100715-P00011
    Figure US20100179803A1-20100715-P00012
    Figure US20100179803A1-20100715-P00013
    Figure US20100179803A1-20100715-P00014
    Figure US20100179803A1-20100715-P00015
    (to female:) you will hear
    Figure US20100179803A1-20100715-P00016
    Figure US20100179803A1-20100715-P00017
    Figure US20100179803A1-20100715-P00018
    Figure US20100179803A1-20100715-P00019
    Figure US20100179803A1-20100715-P00020
    their cars
  • Morphological analysis can be also done based on FSA technology. Linguistic rules can be mostly easily rewritten as FSA rules, allowing an algorithm like the Expectation Maximization algorithm to assign weights to these rules. To allow higher precision for the final translation, the Morphology Analyzer FSA allows to generate multiple hypotheses, leaving the final decision of morphological analysis to a later step. The step of morphological analysis includes also Parts-of-speech tagging, as it is part of the decision for the morphological categories. An example (sub-) FSA for two words out of the examples in Table 2 can be seen in schematics FIG. 4.
  • The syntactic component contains rules annotated with functional expressions and used to build c-structures (hierarchical constituent structures) and their associated f-structures (functional structures) for a sentence.
  • Whenever a rule succeeds, a new edge may be formed; the dag associated with the new complete edge generally contains a CAT (category) label, and an FS (functional structure) label, the value of which will be the unification of the FS labels of the categories that matched in the rule, projected to the functions specified in the rule pattern.
  • Source Sentence Generated Translation
    Figure US20100179803A1-20100715-P00021
    Figure US20100179803A1-20100715-P00022
    Figure US20100179803A1-20100715-P00023
    They will move their cars
  • Constituent Structure
  • Figure US20100179803A1-20100715-P00024
    SBAR
    3
    Figure US20100179803A1-20100715-P00024
    BOSS
    3
    Figure US20100179803A1-20100715-P00024
    VP
    3 3
    Figure US20100179803A1-20100715-P00024
    VBAR
    3 3 3
    Figure US20100179803A1-20100715-P00024
    V
    Figure US20100179803A1-20100715-P00025
    3 3 3 3
    Figure US20100179803A1-20100715-P00024
    VSTEM
    3 3 3 3
    Figure US20100179803A1-20100715-P00024
    mcTNPG
    3 3 3 3
    Figure US20100179803A1-20100715-P00024
    mcVERB-PREFIX
    3 3
    Figure US20100179803A1-20100715-P00024
    NP
    3 3 3
    Figure US20100179803A1-20100715-P00024
    NBAR
    3 3 3 3
    Figure US20100179803A1-20100715-P00024
    N
    Figure US20100179803A1-20100715-P00026
    Figure US20100179803A1-20100715-P00027
    3 3 3 3 3
    Figure US20100179803A1-20100715-P00024
    NSTEM
    3 3 3 3 3
    Figure US20100179803A1-20100715-P00024
    mcNUMBER
    3 3 3 3 3
    Figure US20100179803A1-20100715-P00024
    mcPRON
    3
    Figure US20100179803A1-20100715-P00024
    EOSS
  • Functional Structure
  • Figure US20100179803A1-20100715-P00028
    Figure US20100179803A1-20100715-P00028
    FS [
    Figure US20100179803A1-20100715-P00029
    Figure US20100179803A1-20100715-P00029
    COMPID 103
    Figure US20100179803A1-20100715-P00029
    PRED # SUBJ OBJ1
    Figure US20100179803A1-20100715-P00029
    SUBJ [
    Figure US20100179803A1-20100715-P00030
    Figure US20100179803A1-20100715-P00030
    PERSON THIRD
    Figure US20100179803A1-20100715-P00030
    NUMBER PL
    Figure US20100179803A1-20100715-P00030
    GENDER MSC
    Figure US20100179803A1-20100715-P00030
    FORM PRODROP
    Figure US20100179803A1-20100715-P00030
    Figure US20100179803A1-20100715-P00029
    OBJ1 [
    Figure US20100179803A1-20100715-P00030
    Figure US20100179803A1-20100715-P00030
    CASE ACC
    Figure US20100179803A1-20100715-P00030
    DEFINITE PLUS
    Figure US20100179803A1-20100715-P00030
    POSS [
    Figure US20100179803A1-20100715-P00031
    Figure US20100179803A1-20100715-P00031
    PERSON THIRD
    Figure US20100179803A1-20100715-P00031
    NUMBER PL
    Figure US20100179803A1-20100715-P00031
    GENDER MSC
    Figure US20100179803A1-20100715-P00031
    DEFINITE PLUS
    Figure US20100179803A1-20100715-P00031
    APROFORM
    Figure US20100179803A1-20100715-P00032
    Figure US20100179803A1-20100715-P00031
    Figure US20100179803A1-20100715-P00030
    NUMBER PL
    Figure US20100179803A1-20100715-P00030
    MSALEM MINUS
    Figure US20100179803A1-20100715-P00030
    CONCRETE PLUS
    Figure US20100179803A1-20100715-P00030
    COUNT PLUS
    Figure US20100179803A1-20100715-P00030
    PROPER MINUS
    Figure US20100179803A1-20100715-P00030
    HUMAN MINUS
    Figure US20100179803A1-20100715-P00030
    ANIM MINUS
    Figure US20100179803A1-20100715-P00030
    GENDER FEM
    Figure US20100179803A1-20100715-P00030
    PERSON THIRD
    Figure US20100179803A1-20100715-P00030
    NCLASS NOUN
    Figure US20100179803A1-20100715-P00030
    FORM
    Figure US20100179803A1-20100715-P00033
    Figure US20100179803A1-20100715-P00030
    Figure US20100179803A1-20100715-P00029
    MOOD INDIC
    Figure US20100179803A1-20100715-P00029
    VOICE ACTIVE
    Figure US20100179803A1-20100715-P00029
    AAUXFORM
    Figure US20100179803A1-20100715-P00034
    Figure US20100179803A1-20100715-P00029
    VPREFIX3 VP-S
    Figure US20100179803A1-20100715-P00029
    MODALITY # FUT
    Figure US20100179803A1-20100715-P00029
    GENDER MSC
    Figure US20100179803A1-20100715-P00029
    NUMBER PL
    Figure US20100179803A1-20100715-P00029
    PERSON THIRD
    Figure US20100179803A1-20100715-P00029
    TENSE PRES
    Figure US20100179803A1-20100715-P00029
    ASFGERUND PLUS
    Figure US20100179803A1-20100715-P00029
    VCLASS MAINV
    Figure US20100179803A1-20100715-P00029
    FORM
    Figure US20100179803A1-20100715-P00035
    Figure US20100179803A1-20100715-P00029
    Figure US20100179803A1-20100715-P00028
    TNOD ANY
    Figure US20100179803A1-20100715-P00028
    CAT SBAR
    Figure US20100179803A1-20100715-P00028
  • Debug and Tracing Capabilities of the Parser:
  • a) Trace formation of incomplete edges (each one generates an anticipation for a category)
  • b) Trace formation of complete edges
  • c) Access extended debugs or debug input, such as: Chart, Dictionary, complete edge, List, Morphology, Transfer dag, Quit
  • d) Examine the current status of the chart: check what complete or incomplete edges have been formed at each position in the sentence
  • e) Examine, and optionally save to a file, individual dictionary entries
  • f) Examine, and optionally save to a file, a complete edge's c-structure and f-structure
  • g) Display statistics for a complete edge (number, corresponding incomplete edge number, weight and rule that created it)
  • h) Print a detailed record of the chart operations
  • i) Trace and debug the execution of morphology or syntax rules
  • j) Examine and optionally save to a file a specified level of the transfer and generation DAGS
  • Phrase-Chunking and In-Phrase Re-Ordering
  • The purpose of phrase-chunking is to analyze a sentence and process the recognized phrases in such a way, that the following processes do not destroy the semantic value of the recognized phrase. Within this phrase, a local re-ordering of sub-units (Named entities, words, morphemes) can potentially help improve the translation quality.
  • For example, the (DET-ADJ-N) complex “the big man” in English can have a possible translation with a (DET-N-DET-ADJ) structure in Arabic “
    Figure US20100179803A1-20100715-P00036
    Figure US20100179803A1-20100715-P00037
    , in morphemes “
    Figure US20100179803A1-20100715-P00038
    Figure US20100179803A1-20100715-P00039
    Figure US20100179803A1-20100715-P00040
    , which are very local changes. Phrase-chunking uses linguistically-motivated and hand-crafted rules, as well as structures like statistical N-grams, combined in a statistical FSA, processing multiple hypotheses as an input and generating multiple potential hypotheses.
  • Re-Ordering of the Source Sentence
  • The next step of the translation process is a re-ordering process to make the input more homogeneous in comparison to the data that was used to train the statistical models of the following machine translation steps. To do so, the re-ordering of words is learned statistically by a learning algorithm, which evaluates an automatically generated alignment of a parallel corpus. The result of this algorithm is an FSA, which processes a graph of input tokens (Named Entities, phrases, words, morphemes) and reorders these according to the learned statistical criteria. Re-ordering is a domain specific process, that means for each new task and domain, the re-ordering can be different.
  • Basically, the first automaton of the core translation process is the translation FSA. For this automaton, the training of the statistical lexicon models is initialized using a hand-crafted set of lexical rules. These rules cover lexical information of each individual word as well as they also take some context into account. Naturally, the result of this FSA is not unique, no final decision on selection of words, meanings and translations is being done.
  • Transfer and Translation
  • The main process for the complete translation process is the transfer and translation FSA. For this automaton, the training of the statistical lexicon models is initialized using a hand-crafted set of lexical rules. These rules cover lexical information of each individual word as well as they also take some context into account. Naturally, the result of this FSA is not unique, no final decision on selection of words, meanings and translations is being done.
  • Re-Ordering of the Target Sentence
  • The next step of the translation process is another re-ordering. The goal is to find a word order which is homogenous to the data, which were used to train the statistical language model. To do so, again the re-ordering of words is learned statistically by a learning algorithm, which evaluates the order of the language model corpus. The result of this algorithm is an FSA, which processes a graph of input tokens (Named Entities, phrases, words, morphemes) and reorders these according to the learned statistical criteria. Also here, the re-ordering is a domain specific process, that means for each new task and domain, the re-ordering can be different and needs to be developed specific to the task.
  • Language Model Re-Scoring
  • The last step in the process is to select the best translation of the generated graph, calculating the weights and scores of each of the previous processing steps and combining these scores with the language model score. The language model score results off of a combination of different sub-language models, namely n-gram as well as grammatical parsers, as well as gap-n-gram language models. The combination is learned using a maximum entropy approach.
  • As (Och&Ney, 2004) described, if used in a Maximum Entropy framework, by using:
  • Pr ( e f ) = p l 1 M ( e f ) = exp ( m = 1 M l m h m ( e , f ) ) e exp ( m = 1 M l m h m ( e , f ) ) ;
  • The lexical entries can be described as follows:
  • h TM x ( f , e ) = ( j = 1 J δ ( f j , f ) ) · ( i = 1 I δ ( e i , e ) )
  • where the statistically learned lexical entries are in the feature function hTM s (f,e), manually generated lexical rules are in the feature function hTM r (f,e).
  • The lexical entries can be described as follows:
  • h LM x ( e ) = i = 1 I h LM x ( e i )
  • The use of functional constraints for lexical information in source and target give a deeper syntactic and semantic analytic value to the translation. Functional constraints are multiple, and some of these functions are language dependent (e.g. gender, polarity, mood, etc.). The use of these functional constraints can be denoted as: hFM x (f,e)=zx(f,e); where the functions in use are denoted by z(.). Some of these functions are sentence related, so the functional feature can be defined as: hFM x (f,e)=δ(z(f),z(e)); but most functional relations are word based.
  • These functions can be cross-language or within a certain language. A cross-language function could be the tense information but also the function “human”, describing that the concept to be generally a human being (“man”, “woman”, “president” are generally “human”, but also potentially concepts like “manager”, “caller”, “driver”, depending on the semantic and syntactic environment). Such a functional feature can be mathematically defined in the following way:
  • h FM x ( f , e ) = j = 1 J δ ( z ( f j ) , z ( f ) ) · · i = 1 I δ ( z ( e i ) , z ( e ) )
  • A “within-language” function could be gender, as objects can have different genders in different languages (e.g. for the translation eqivalents of “table”, “table” in English has no gender, “Tisch” in German is masculine, “table” in French is feminine, “ ” in Arabic is feminine). The following is an example, which affects only the generation of the target language:
  • h FM x ( f , e ) = j = 1 J k = j J δ ( z ( e j ) , z ( e k ) )
  • Speech Input Processing
  • The translation search is a function, which takes the output of all the above mentioned models and takes the probability of the different subtopics and calculates the path of the best hypothesis, traversing through the sub-FSAs described in the previous chapter.
  • Combining speech recognition output with machine translation is relatively straightforward. Speech recognition output for machine translation can be coded in word lattices to allow for multiple hypotheses (Vogel et al. 2000, Khadivi et al. 2006). Using FSAs allow for graphs (word lattices or confusion networks) to be an input instead of single-best word sequences, so that potential errors can be recovered. This is the case, as the speech recognition error rate on the first-best result is in general worse than the result encoded in a word lattice, which is a very efficient way to save a n-best list of hypotheses.
  • Component Integration and Optimization
  • The weight of each of the different components which were described in this chapter are being learned using the Minimum-Error rate-Training (MET). The sub-language models are dealt with as own units in the MET and optimized on this level.
  • TABLE 2 Corpus statistics about the Iraqi Colloquial Arabic and Arabic Broadcast News ICA (Telephone) MSA (Broadcast/News) Vocabulary (full-form) 65,000 400,000 Running parallel text 500,000 260,000,000 Running monolingual text 4,000,000 5,000,000,000 Running target text 4,800,000 8,000,000,000 OOV Rate 1% <0.1%
  • Customization and Domain-Adaptation
  • To increase quality of the hybrid machine translation, similar to purely statistical machine translation systems, most components can be adapted to new environments, either by using monolingual data (for language mode training), bilingual data (for lexicon models, alignment models and alignment templates) as well as linguistic information, like lexical and functional lexicons, syntactic rules for parsing and/or transfer.
  • Statistical MT Module
  • Statistical Machine Translation (SMT) systems have the advantage of being able to learn translations of phrases, not just individual words, which permits them to improve the functionality of both example-based approaches and translation memory. Another advantage to some SMT systems, which use the Maximum Entropy approach, is that they combine many knowledge sources and, therefore give a good basis for making use of multiple knowledge sources while analyzing a sentence for translation. In general, fully functional statistical machine translation systems exist and may be implemented as part of a hybrid system as defined herein. One example is presented in U.S. Pat. No. 5,991,710 where is incorporated by reference herein.
  • Rule-Based MT Module
  • A rule-based module, an Lexical Functional Grammar (“LFG”) system (Shihadah & Roochnik, 1998) may be employed which is used to feed the hybrid machine translation engine. The LFG system contains a richly-annotated lexicon containing functional and semantic information. There are many examples of fully functional rule based machine translation modules and systems which may be used as part of the hybrid machine translation system described herein, including the one shown in U.S. Pat. No. 6,952,665 which is incorporated by reference herein.
  • Hybrid MT
  • In the hybrid machine translation (HMT) framework, the statistical search process may have full access to the information available in LFG lexical entries, grammatical rules, constituent structures and functional structures. This is accomplished by treating the pieces of information as feature functions in the Maximum Entropy algorithm of the hybrid machine translation engine shown in FIG. 1, which is also shown as the decoder in FIGS. 6A-C.
  • Incorporation of these knowledge sources both expand and constrain the search possibilities. Areas where the search is expanded include those in which the two languages differ significantly, as for example when a long-distance dependency exists in one language but not the other.
  • Statistical Machine Translation is traditionally represented in the literature as choosing the target (e.g., English) sentence with the highest probabil-ity given a source (e.g., French) sentence. Originally, and most-commonly, SMT uses the “noisy channel” or “source-channel” model adapted from speech recognition (Brown et. al., 1990; Brown et. al., 1993).
  • While most SMT systems used to be based on the traditional “noisy channel” approach, this is simply one method of composing a decision rule that determines the best translation. Other methods may be employed and many of them can even be combined if a direct translation model using a Maximum Entropy is employed.
  • Translation Models
  • The translation models introduced for the system which is described herein is a combination of statistically learned lexicons interpolated with a bilingual lexicon used in the rule-based LFG system.
  • Language Models
  • The use of lexical (syntactic and semantic) and grammatical models or feature functions in a statistical framework is introduced. The incorporation of rich lexical and structural data into SMT allows the application of linguistic analysis to SMT. To improve MT quality language model feature functions, the language model feature functions may cover standard 5-gram, POS-based 5-gram and time-synchronous CYK-type parser, as described in (Sawaf et. al., 2000). The m-gram language models (word and POS class-based) may be trained on a corpus, where morphological analysis is utilized.
  • Then a hybrid translation system is trained to translate the large training corpus in non-dialect language into the targeted dialect language. After that, the new “artificially” generated corpus is utilized to train the statistical language models. For the words, which do not have a translation into the target language, may be transliterated, using a transliteration engine, for example one based on the Grapheme-to-Phoneme converter like (Bisa-ni & Ney, 2002; Wei, 2004). Besides this corpus, the original text corpus is used for the training of the final language models.
  • Functional Models
  • The use of functional constraints for lexical information in source and target give a deeper syntactic and semantic analytic value to the translation. Functional constraints are multiple, and some of these functions are language dependent (e.g. gender, polarity, mood, etc.).
  • These functions can be cross-language or within a certain language. A cross-language function could be the tense information but also the function “human”, describing that the concept to be generally a human being (“man”, “woman”, “president” are generally “human”, but also po-tentially concepts like “manager”, “caller”, “driv-er”, depending on the semantic and syntactic en-vironment).
  • A “within-language” function could be gender, as objects can have different genders in different languages (e.g. for the translation equivalents of “table”, in English it has no gender, in German it is masculine, in French and Arabic it is feminine).
  • Translation of Standardized Foreign Languages into English
  • The translation from standardized languages (written text not from dialects as in chat, IM and email) into English is done using the above de-scribed system using lexical, functional and syn-tactical features which were used in a LFG based system. The statistical models are trained on a bi-lingual sentence aligned corpus for the translation model and the alignment template model. The language models (POS-based and word-based) are being trained on a monolingual corpus.
  • Translation of Dialect into English
  • Translation of foreign dialect may be implemented in one embodiment using a hybrid MT system that translates the dialect into the written standard of the foreign language first. For the presented translation system, a bilingual corpus is used which consists of sentences in the dialect and the standardized form for training the language models. Also feature functions built out of rules built to translate (or rather: convert) the dialect into non-dialect are used.
  • As much of the input can be either standardized or dialect at the same time, the quality of translation can be increased by using dialect feature functions for both the standardized and dialect variants and allow the Generative Iterative Scaling (GIS) algorithm to change the weighting of these features during the training process.
  • In addition to the challenge of dialects, are the nuances of various topical areas (domains) that present unique terms specific to those areas or commonly used terms with different meanings specific to those areas. For example, a text might report on the year's flu season and then a malicious attack spread by a hacker. Both topics might use the term “virus” with different meanings. The subsequent translation could be inaccurate with-out the proper understanding of the context of the term “virus” in each instance. The use of domain-specific information resolves this potential prob-lem.
  • Domains and Micro-Dictionaries
  • The introduction of special domain dictionaries is readily available. Multiple domain specific on-line dictionaries include, in addition to the general dictionary, for example the following micro-dictionaries:
      • Military;
      • Special Operations;
      • Mechanical;
      • Political & Diplomatic;
      • Nuclear;
      • Chemical;
      • Aviation;
      • Computer & Technology;
      • Medical;
      • Business & Economics;
      • Law Enforcement;
      • Drug terms
  • FIGS. 6A, 6B and 6C depict other views of a hybrid machine translation system according to an embodiment of the invention. As is apparent from FIGS. 6A-6C, the different rule based and statistical components provide outputs to a decoder which may implement a maximum entropy algorithm and may take into account these weights when selecting the optimum translation.
  • FIG. 4 shows a process for creating an annotated DAG. Referring to FIG. 4, in step 400 a DAG is created. This may be performed either by a statistical machine translation module or a rule base machine translation module. In the next step, 410, rule based language models, such as lexical models comprising syntactic or semantic models, rules or constraints or functionalities are applied to or related to the DAG, or in other words are used to annotate the DAG. The result is that certain paths within the DAG are excluded or at least made more improbable. Finally, the annotated DAG is used in step 420 as a basis for translation by the hybrid machine translation engine using a maximum entropy model that receives the DAG as an input along with other inputs from statistical machine translation modules and rule based models, such as from the models shown in FIG. 1.
  • FIG. 5 shows a hybrid machine translation engine 500 according to an embodiment of the invention. It may be used to process source text 510 directly or to process lattices of text output from a speech recognition processor 520. A hybrid machine translation engine 530 may receive source text from both types of sources and produce target text in a different language according to the hybrid translation processes and systems as described herein to produce target text 540.
  • The processes described herein, and with reference to the Figures may all be implemented by a general purpose computer or server having a processor, a memory and input/output capabilities including any or all of a display, a keyboard, a network interface, a database interface. The network interfaces may be electrical, optical, wireless or any other suitable technology. And, there may be a single computer which performs the processes or multiple computers at different locations that share or pass data between and among themselves over a network, for example. The various models and data used by the language translation process including databases, the text, annotated text or recognized text processed may be stored in memory either locally on the computer in, for example, RAM or on a hard drive or other computer usable media, or may be stored on and be accessible over a network. The processes may also be implement using computer instructions that are stored in memory and executed by the computer processor to perform the steps and transform the input text or speech to translated text or speech.
  • While particular embodiments of the present invention have been shown and described, it will be understood by those having ordinary skill in the art that changes may be made to those embodiments without departing from the spirit and scope of the present invention.

Claims (19)

1. A system for automatic translation, comprising:
at least one database for storing source and target text and rule based and statistical language models;
a rule based machine translation module that translates source text based on the rule based models;
a statistical machine translation module based on maximum entropy modeling that translates the source text based on statistical language models;
a hybrid machine translation engine, having a maximum entropy algorithm that is coupled to the rule based machine translation module and the statistical machine translation module that is capable of translating source text into target text based on the rule based and statistical language models.
2. The system according to claim 1, wherein the maximum entropy algorithm of the hybrid machine translation engine is configured to assign probabilities to different statistical language models and different rule based models.
3. The system according to claim 3, wherein the hybrid machine translation engine operates to perform translation of source text into target text based on available models.
4. The system according to claim 4, wherein the hybrid machine translation engine operates to perform translation of source text into target text when there are no statistical models available.
5. The system according to claim 4, wherein the hybrid machine translation engine operates to perform translation of source text into target text when there are no rule based models available.
6. The system according to claim 1, wherein the database further stores directed acyclic graphs (DAGS) used by both the rule based and statistical translation systems.
7. The system according to claim 1, wherein the rule based translation system includes syntactic, semantic and grammatical based rules.
8. The system according to claim 6, wherein the rule based translation system includes syntactic, semantic and grammatical based rules, wherein at least one of the semantic rules are used to eliminate paths from the DAGS used by the hybrid machine translation engine.
9. They system according to claim 6, wherein the DAGS are created by the statistical translation system and the rule based translation system includes semantic rules that are used to annotate the statistical DAGS.
10. The system according to claim 9, wherein the hybrid machine translation engine uses the annotated statistical DAGS for translation of the source text into target text.
11. A system for automatic translation of source text into target text, comprising:
at least one database for storing source and target text and directed acyclic graphs (DAGS) and semantic rules;
a statistical translation system that generates DAGS based on source text;
a hybrid machine translation engine that annotates the DAGS based on semantic rules and translates source text into target text based on at least the DAGS.
12. The system according to claim 11, wherein the hybrid machine translation engine uses a maximum entropy algorithm to perform the translation.
13. A method for machine translation comprising:
receiving and storing source text;
generating directed acyclic graphs (DAGS) based on statistical models;
annotating the DAGS based on semantic rules; and
translating the source text into target text based at least in part on the annotated DAGS.
14. The method according to claim 13, wherein the translating is performed based on a maximum entropy algorithm.
15. The method according to claim 14, wherein the hybrid machine translation engine further receives weighted inputs from statistical and rule based translation system models and performs the translating based on the weighted inputs.
16. The method according to claim 15, wherein the statistical models include a statistical lexicon model, a statistical phrase model and a statistical language model.
17. The method according to claim 15, wherein the rule based models include a rule based syntactic model, a rule based semantic model, a rule based grammatical model and a rule based transfer system model.
18. The method according to claim 14, wherein the hybrid machine translation engine further receives weighted inputs from a statistical lexicon model, a statistical phrase model, a statistical language model, a rule based syntactic model, a rule based semantic model, a rule based grammatical model and a rule based transfer system model.
19. The method according to claim 14, wherein the hybrid machine translation engine further receives weighted inputs from at least four of: a statistical lexicon model, a statistical phrase model, a statistical language model, a rule based syntactic model, a rule based semantic model, a rule based grammatical model and a rule based transfer system model.
US12/606,110 2008-10-24 2009-10-26 Hybrid machine translation Active 2033-09-08 US9798720B2 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US19305808P true 2008-10-24 2008-10-24
US12/606,110 US9798720B2 (en) 2008-10-24 2009-10-26 Hybrid machine translation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US12/606,110 US9798720B2 (en) 2008-10-24 2009-10-26 Hybrid machine translation

Publications (2)

Publication Number Publication Date
US20100179803A1 true US20100179803A1 (en) 2010-07-15
US9798720B2 US9798720B2 (en) 2017-10-24

Family

ID=42119761

Family Applications (1)

Application Number Title Priority Date Filing Date
US12/606,110 Active 2033-09-08 US9798720B2 (en) 2008-10-24 2009-10-26 Hybrid machine translation

Country Status (2)

Country Link
US (1) US9798720B2 (en)
WO (1) WO2010046782A2 (en)

Cited By (60)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100228538A1 (en) * 2009-03-03 2010-09-09 Yamada John A Computational linguistic systems and methods
US20110282643A1 (en) * 2010-05-11 2011-11-17 Xerox Corporation Statistical machine translation employing efficient parameter training
US20110307244A1 (en) * 2010-06-11 2011-12-15 Microsoft Corporation Joint optimization for machine translation system combination
US20120016657A1 (en) * 2010-07-13 2012-01-19 Dublin City University Method of and a system for translation
US20120150529A1 (en) * 2010-12-09 2012-06-14 Electronics And Telecommunication Research Institute Method and apparatus for generating translation knowledge server
US20120166183A1 (en) * 2009-09-04 2012-06-28 David Suendermann System and method for the localization of statistical classifiers based on machine translation
US20120209590A1 (en) * 2011-02-16 2012-08-16 International Business Machines Corporation Translated sentence quality estimation
US8296142B2 (en) 2011-01-21 2012-10-23 Google Inc. Speech recognition using dock context
US8352246B1 (en) 2010-12-30 2013-01-08 Google Inc. Adjusting language models
US20130054224A1 (en) * 2011-08-30 2013-02-28 Dublin City University Method and system for enhancing text alignment between a source language and a target language during statistical machine translation
US20130226556A1 (en) * 2010-11-05 2013-08-29 Sk Planet Co., Ltd. Machine translation device and machine translation method in which a syntax conversion model and a word translation model are combined
US20130262079A1 (en) * 2012-04-03 2013-10-03 Lindsay D'Penha Machine language interpretation assistance for human language interpretation
US8583432B1 (en) 2012-07-18 2013-11-12 International Business Machines Corporation Dialect-specific acoustic language modeling and speech recognition
US20140039879A1 (en) * 2011-04-27 2014-02-06 Vadim BERMAN Generic system for linguistic analysis and transformation
US20140067361A1 (en) * 2012-08-28 2014-03-06 Xerox Corporation Lexical and phrasal feature domain adaptation in statistical machine translation
US8751217B2 (en) 2009-12-23 2014-06-10 Google Inc. Multi-modal input on an electronic device
US8818792B2 (en) * 2010-11-05 2014-08-26 Sk Planet Co., Ltd. Apparatus and method for constructing verbal phrase translation pattern using bilingual parallel corpus
US8838434B1 (en) * 2011-07-29 2014-09-16 Nuance Communications, Inc. Bootstrap call router to other languages using selected N-best translations
WO2014144716A1 (en) * 2013-03-15 2014-09-18 The Dun & Bradstreet Corporation Multi-lingual business indicia curation and transliteration synthesis
US20150066503A1 (en) * 2013-08-28 2015-03-05 Verint Systems Ltd. System and Method of Automated Language Model Adaptation
US20150066502A1 (en) * 2013-08-28 2015-03-05 Verint Systems Ltd. System and Method of Automated Model Adaptation
WO2013123384A3 (en) * 2012-02-16 2015-06-18 Standingdeer John C Deconstruction and construction of words of a polysynthetic language for translation purposes
US20150178274A1 (en) * 2013-12-25 2015-06-25 Kabushiki Kaisha Toshiba Speech translation apparatus and speech translation method
US20150228272A1 (en) * 2014-02-08 2015-08-13 Honda Motor Co., Ltd. Method and system for the correction-centric detection of critical speech recognition errors in spoken short messages
US20150302001A1 (en) * 2012-02-16 2015-10-22 Continental Automotive Gmbh Method and device for phonetizing data sets containing text
US9412365B2 (en) 2014-03-24 2016-08-09 Google Inc. Enhanced maximum entropy models
EP3062238A1 (en) * 2015-02-27 2016-08-31 Samsung Electronics Co., Ltd. Summarization by sentence extraction and translation of summaries containing named entities
US9530161B2 (en) 2014-02-28 2016-12-27 Ebay Inc. Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US9569526B2 (en) 2014-02-28 2017-02-14 Ebay Inc. Automatic machine translation using user feedback
US20170060854A1 (en) * 2015-08-25 2017-03-02 Alibaba Group Holding Limited Statistics-based machine translation method, apparatus and electronic device
US20170185588A1 (en) * 2015-12-28 2017-06-29 Facebook, Inc. Predicting future translations
US9779724B2 (en) 2013-11-04 2017-10-03 Google Inc. Selecting alternates in speech recognition
US9798720B2 (en) 2008-10-24 2017-10-24 Ebay Inc. Hybrid machine translation
US9842592B2 (en) 2014-02-12 2017-12-12 Google Inc. Language models using non-linguistic context
US9881006B2 (en) 2014-02-28 2018-01-30 Paypal, Inc. Methods for automatic generation of parallel corpora
US9916306B2 (en) 2012-10-19 2018-03-13 Sdl Inc. Statistical linguistic analysis of source content
US9940658B2 (en) 2014-02-28 2018-04-10 Paypal, Inc. Cross border transaction machine translation
US9954794B2 (en) 2001-01-18 2018-04-24 Sdl Inc. Globalization management system and method therefor
US9978367B2 (en) 2016-03-16 2018-05-22 Google Llc Determining dialog states for language models
US9984054B2 (en) 2011-08-24 2018-05-29 Sdl Inc. Web interface including the review and manipulation of a web document and utilizing permission based control
US10013417B2 (en) 2014-06-11 2018-07-03 Facebook, Inc. Classifying languages for objects and entities
US10061749B2 (en) 2011-01-29 2018-08-28 Sdl Netherlands B.V. Systems and methods for contextual vocabularies and customer segmentation
US10089299B2 (en) 2015-12-17 2018-10-02 Facebook, Inc. Multi-media context language processing
US10134394B2 (en) 2015-03-20 2018-11-20 Google Llc Speech recognition using log-linear model
US10140320B2 (en) 2011-02-28 2018-11-27 Sdl Inc. Systems, methods, and media for generating analytical data
US20180373704A1 (en) * 2017-06-21 2018-12-27 Samsung Electronics Co., Ltd. Method and apparatus for machine translation using neural network and method of training the apparatus
US10185763B2 (en) * 2016-11-30 2019-01-22 Facebook, Inc. Syntactic models for parsing search queries on online social networks
US10198438B2 (en) 1999-09-17 2019-02-05 Sdl Inc. E-services translation utilizing machine translation and translation memory
US10248650B2 (en) 2004-03-05 2019-04-02 Sdl Inc. In-context exact (ICE) matching
US10255275B2 (en) 2015-08-25 2019-04-09 Alibaba Group Holding Limited Method and system for generation of candidate translations
US10261994B2 (en) 2012-05-25 2019-04-16 Sdl Inc. Method and system for automatic management of reputation of translators
US10311860B2 (en) 2017-02-14 2019-06-04 Google Llc Language model biasing system
US10319252B2 (en) 2005-11-09 2019-06-11 Sdl Inc. Language capability assessment and training apparatus and techniques
US10346537B2 (en) 2015-09-22 2019-07-09 Facebook, Inc. Universal translation
US10380249B2 (en) 2017-10-02 2019-08-13 Facebook, Inc. Predicting future trending topics
US10409913B2 (en) 2015-10-01 2019-09-10 Conduent Business Services, Llc Methods and systems to train classification models to classify conversations
US10417646B2 (en) 2010-03-09 2019-09-17 Sdl Inc. Predicting the cost associated with translating textual content
US10452740B2 (en) 2012-09-14 2019-10-22 Sdl Netherlands B.V. External content libraries
US10452786B2 (en) 2014-12-29 2019-10-22 Paypal, Inc. Use of statistical flow data for machine translations between different languages
US10540450B2 (en) 2017-09-05 2020-01-21 Facebook, Inc. Predicting future translations

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150039286A1 (en) * 2013-07-31 2015-02-05 Xerox Corporation Terminology verification systems and methods for machine translation services for domain-specific texts

Citations (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5912986A (en) * 1994-06-21 1999-06-15 Eastman Kodak Company Evidential confidence measure and rejection technique for use in a neural network based optical character recognition system
US6028956A (en) * 1997-04-04 2000-02-22 Kofile Inc. Object location and span determination method and apparatus which determines a location and span of an object in an image
US6490549B1 (en) * 2000-03-30 2002-12-03 Scansoft, Inc. Automatic orthographic transformation of a text stream
US20020198713A1 (en) * 1999-01-29 2002-12-26 Franz Alexander M. Method and apparatus for perfoming spoken language translation
US20040181410A1 (en) * 2003-03-13 2004-09-16 Microsoft Corporation Modelling and processing filled pauses and noises in speech recognition
US20050015217A1 (en) * 2001-11-16 2005-01-20 Galia Weidl Analyzing events
US6865528B1 (en) * 2000-06-01 2005-03-08 Microsoft Corporation Use of a unified language model
US20050228640A1 (en) * 2004-03-30 2005-10-13 Microsoft Corporation Statistical language model for logical forms
US20060142995A1 (en) * 2004-10-12 2006-06-29 Kevin Knight Training for a text-to-text application which uses string to tree conversion for training and decoding
US20060241869A1 (en) * 2003-08-05 2006-10-26 Rosetta Inpharmatics Llc Computer systems and methods for inferring causality from cellullar constituent abundance data
US20070166707A1 (en) * 2002-12-27 2007-07-19 Rosetta Inpharmatics Llc Computer systems and methods for associating genes with traits using cross species data
US20070185946A1 (en) * 2004-02-17 2007-08-09 Ronen Basri Method and apparatus for matching portions of input images
US20070209075A1 (en) * 2006-03-04 2007-09-06 Coffman Thayne R Enabling network intrusion detection by representing network activity in graphical form utilizing distributed data sensors to detect and transmit activity data
US7269598B2 (en) * 2000-03-22 2007-09-11 Insightful Corporation Extended functionality for an inverse inference engine based web search
US20080025617A1 (en) * 2006-07-25 2008-01-31 Battelle Memorial Institute Methods and apparatuses for cross-ontologial analytics
US20080040095A1 (en) * 2004-04-06 2008-02-14 Indian Institute Of Technology And Ministry Of Communication And Information Technology System for Multiligual Machine Translation from English to Hindi and Other Indian Languages Using Pseudo-Interlingua and Hybridized Approach
US20080077391A1 (en) * 2006-09-22 2008-03-27 Kabushiki Kaisha Toshiba Method, apparatus, and computer program product for machine translation
US20080114581A1 (en) * 2006-11-15 2008-05-15 Gil Meir Root cause analysis approach with candidate elimination using network virtualization
US20080133245A1 (en) * 2006-12-04 2008-06-05 Sehda, Inc. Methods for speech-to-speech translation
US20080154577A1 (en) * 2006-12-26 2008-06-26 Sehda,Inc. Chunk-based statistical machine translation system
US20080270109A1 (en) * 2004-04-16 2008-10-30 University Of Southern California Method and System for Translating Information with a Higher Probability of a Correct Translation
US20090018821A1 (en) * 2006-02-27 2009-01-15 Nec Corporation Language processing device, language processing method, and language processing program
US20090070099A1 (en) * 2006-10-10 2009-03-12 Konstantin Anisimovich Method for translating documents from one language into another using a database of translations, a terminology dictionary, a translation dictionary, and a machine translation system
US20090157380A1 (en) * 2007-12-18 2009-06-18 Electronics And Telecommunications Research Institute Method and apparatus for providing hybrid automatic translation
US20090265230A1 (en) * 2008-04-18 2009-10-22 Yahoo! Inc. Ranking using word overlap and correlation features
US20090265290A1 (en) * 2008-04-18 2009-10-22 Yahoo! Inc. Optimizing ranking functions using click data
US7610189B2 (en) * 2001-10-18 2009-10-27 Nuance Communications, Inc. Method and apparatus for efficient segmentation of compound words using probabilistic breakpoint traversal
US20100004920A1 (en) * 2008-07-03 2010-01-07 Google Inc. Optimizing parameters for machine translation
US7716226B2 (en) * 2005-09-27 2010-05-11 Patentratings, Llc Method and system for probabilistically quantifying and visualizing relevance between two or more citationally or contextually related data objects
US7716039B1 (en) * 2006-04-14 2010-05-11 At&T Intellectual Property Ii, L.P. Learning edit machines for robust multimodal understanding
US20100180244A1 (en) * 2009-01-15 2010-07-15 International Business Machines Corporation Method For Efficiently Checkpointing And Restarting Static Timing Analysis Of An Integrated Circuit Chip

Family Cites Families (54)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5175684A (en) 1990-12-31 1992-12-29 Trans-Link International Corp. Automatic text translation and routing system
JP2745370B2 (en) 1993-02-23 1998-04-28 日本アイ・ビー・エム株式会社 Machine translation method and machine translation apparatus
JPH08190562A (en) 1995-01-06 1996-07-23 Ricoh Co Ltd Machine translation system
JP2003529845A (en) 2000-03-31 2003-10-07 アミカイ・インコーポレイテッド Method and apparatus for providing multilingual translation over a network
US8396859B2 (en) 2000-06-26 2013-03-12 Oracle International Corporation Subject matter context search engine
US6885985B2 (en) 2000-12-18 2005-04-26 Xerox Corporation Terminology translation for unaligned comparable corpora using category based translation probabilities
US7054803B2 (en) 2000-12-19 2006-05-30 Xerox Corporation Extracting sentence translations from translated documents
GB2390704A (en) 2002-07-09 2004-01-14 Canon Kk Automatic summary generation and display
JP2004220266A (en) 2003-01-14 2004-08-05 Cross Language Inc Machine translation device and machine translation method
US7139752B2 (en) 2003-05-30 2006-11-21 International Business Machines Corporation System, method and computer program product for performing unstructured information management and automatic text analysis, and providing multiple document views derived from different document tokenizations
US20050005237A1 (en) 2003-07-03 2005-01-06 Rail Peter D. Method for maintaining a centralized, multidimensional master index of documents from independent repositories
US7792884B2 (en) 2005-05-26 2010-09-07 Itelehealth, Inc. System and method for conducting tailored search
EP1969493B1 (en) 2006-02-17 2018-12-26 Google LLC Encoding and adaptive, scalable accessing of distributed models
US20100280818A1 (en) 2006-03-03 2010-11-04 Childers Stephen R Key Talk
US7542893B2 (en) 2006-05-10 2009-06-02 Xerox Corporation Machine translation using elastic chunks
US7870117B1 (en) 2006-06-01 2011-01-11 Monster Worldwide, Inc. Constructing a search query to execute a contextual personalized search of a knowledge base
US7805289B2 (en) 2006-07-10 2010-09-28 Microsoft Corporation Aligning hierarchal and sequential document trees to identify parallel data
US20080077384A1 (en) 2006-09-22 2008-03-27 International Business Machines Corporation Dynamically translating a software application to a user selected target language that is not natively provided by the software application
US7774193B2 (en) 2006-12-05 2010-08-10 Microsoft Corporation Proofing of word collocation errors based on a comparison with collocations in a corpus
US7983897B2 (en) 2007-02-14 2011-07-19 Google Inc. Machine translation feedback
US7949514B2 (en) 2007-04-20 2011-05-24 Xerox Corporation Method for building parallel corpora
WO2009038525A1 (en) 2007-09-17 2009-03-26 Capfinder Aktiebolag System for assisting in drafting applications
JP2009075795A (en) 2007-09-20 2009-04-09 National Institute Of Information & Communication Technology Machine translation device, machine translation method, and program
WO2009039524A1 (en) 2007-09-21 2009-03-26 Google Inc. Cross-language search
JP5288371B2 (en) 2008-06-03 2013-09-11 独立行政法人情報通信研究機構 Statistical machine translation system
KR100961717B1 (en) 2008-09-16 2010-06-10 한국전자통신연구원 Method and apparatus for detecting errors of machine translation using parallel corpus
US8775154B2 (en) 2008-09-18 2014-07-08 Xerox Corporation Query translation through dictionary adaptation
WO2010036287A1 (en) 2008-09-24 2010-04-01 Pacific Biosciences Of California, Inc. Intermittent detection during analytical reactions
WO2010046782A2 (en) 2008-10-24 2010-04-29 App Tek Hybrid machine translation
US8306806B2 (en) 2008-12-02 2012-11-06 Microsoft Corporation Adaptive web mining of bilingual lexicon
US8244519B2 (en) 2008-12-03 2012-08-14 Xerox Corporation Dynamic translation memory using statistical machine translation
US8285536B1 (en) 2009-07-31 2012-10-09 Google Inc. Optimizing parameters for machine translation
US9053202B2 (en) 2009-09-25 2015-06-09 Yahoo! Inc. Apparatus and methods for user generated translation
US8380486B2 (en) 2009-10-01 2013-02-19 Language Weaver, Inc. Providing machine-generated translations and corresponding trust levels
US8930176B2 (en) 2010-04-01 2015-01-06 Microsoft Corporation Interactive multilingual word-alignment techniques
EP2593884A2 (en) 2010-07-13 2013-05-22 Motionpoint Corporation Dynamic language translation of web site content
US20120109623A1 (en) 2010-11-01 2012-05-03 Microsoft Corporation Stimulus Description Collections
US20120141959A1 (en) 2010-12-07 2012-06-07 Carnegie Mellon University Crowd-sourcing the performance of tasks through online education
US8645289B2 (en) 2010-12-16 2014-02-04 Microsoft Corporation Structured cross-lingual relevance feedback for enhancing search results
US8650023B2 (en) 2011-03-21 2014-02-11 Xerox Corporation Customer review authoring assistant
US8732151B2 (en) 2011-04-01 2014-05-20 Microsoft Corporation Enhanced query rewriting through statistical machine translation
US20120290487A1 (en) 2011-04-15 2012-11-15 IP Street Evaluating intellectual property
US8620931B2 (en) 2011-06-24 2013-12-31 Siemens Aktiengesellschaft Method of composing an ontology alignment
US8781810B2 (en) 2011-07-25 2014-07-15 Xerox Corporation System and method for productive generation of compound words in statistical machine translation
US9471565B2 (en) 2011-07-29 2016-10-18 At&T Intellectual Property I, L.P. System and method for locating bilingual web sites
US20130110596A1 (en) 2011-10-19 2013-05-02 Article One Partners Holdings Foreign language incentive structure for crowdsourced research
US8386477B1 (en) 2011-11-30 2013-02-26 Google Inc. System and method for determining user language intent
US9323746B2 (en) 2011-12-06 2016-04-26 At&T Intellectual Property I, L.P. System and method for collaborative language translation
US9342503B1 (en) 2013-03-12 2016-05-17 Amazon Technologies, Inc. Correlation across languages
US9195654B2 (en) 2013-03-15 2015-11-24 Google Inc. Automatic invocation of a dialog user interface for translation applications
US9530161B2 (en) 2014-02-28 2016-12-27 Ebay Inc. Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US9569526B2 (en) 2014-02-28 2017-02-14 Ebay Inc. Automatic machine translation using user feedback
US9881006B2 (en) 2014-02-28 2018-01-30 Paypal, Inc. Methods for automatic generation of parallel corpora
US9940658B2 (en) 2014-02-28 2018-04-10 Paypal, Inc. Cross border transaction machine translation

Patent Citations (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5912986A (en) * 1994-06-21 1999-06-15 Eastman Kodak Company Evidential confidence measure and rejection technique for use in a neural network based optical character recognition system
US6028956A (en) * 1997-04-04 2000-02-22 Kofile Inc. Object location and span determination method and apparatus which determines a location and span of an object in an image
US20020198713A1 (en) * 1999-01-29 2002-12-26 Franz Alexander M. Method and apparatus for perfoming spoken language translation
US7269598B2 (en) * 2000-03-22 2007-09-11 Insightful Corporation Extended functionality for an inverse inference engine based web search
US6490549B1 (en) * 2000-03-30 2002-12-03 Scansoft, Inc. Automatic orthographic transformation of a text stream
US6865528B1 (en) * 2000-06-01 2005-03-08 Microsoft Corporation Use of a unified language model
US7610189B2 (en) * 2001-10-18 2009-10-27 Nuance Communications, Inc. Method and apparatus for efficient segmentation of compound words using probabilistic breakpoint traversal
US20050015217A1 (en) * 2001-11-16 2005-01-20 Galia Weidl Analyzing events
US20070166707A1 (en) * 2002-12-27 2007-07-19 Rosetta Inpharmatics Llc Computer systems and methods for associating genes with traits using cross species data
US20040181410A1 (en) * 2003-03-13 2004-09-16 Microsoft Corporation Modelling and processing filled pauses and noises in speech recognition
US20060241869A1 (en) * 2003-08-05 2006-10-26 Rosetta Inpharmatics Llc Computer systems and methods for inferring causality from cellullar constituent abundance data
US20070038386A1 (en) * 2003-08-05 2007-02-15 Schadt Eric E Computer systems and methods for inferring casuality from cellular constituent abundance data
US20070185946A1 (en) * 2004-02-17 2007-08-09 Ronen Basri Method and apparatus for matching portions of input images
US20050228640A1 (en) * 2004-03-30 2005-10-13 Microsoft Corporation Statistical language model for logical forms
US20080040095A1 (en) * 2004-04-06 2008-02-14 Indian Institute Of Technology And Ministry Of Communication And Information Technology System for Multiligual Machine Translation from English to Hindi and Other Indian Languages Using Pseudo-Interlingua and Hybridized Approach
US20080270109A1 (en) * 2004-04-16 2008-10-30 University Of Southern California Method and System for Translating Information with a Higher Probability of a Correct Translation
US20060142995A1 (en) * 2004-10-12 2006-06-29 Kevin Knight Training for a text-to-text application which uses string to tree conversion for training and decoding
US7716226B2 (en) * 2005-09-27 2010-05-11 Patentratings, Llc Method and system for probabilistically quantifying and visualizing relevance between two or more citationally or contextually related data objects
US20090018821A1 (en) * 2006-02-27 2009-01-15 Nec Corporation Language processing device, language processing method, and language processing program
US20070209075A1 (en) * 2006-03-04 2007-09-06 Coffman Thayne R Enabling network intrusion detection by representing network activity in graphical form utilizing distributed data sensors to detect and transmit activity data
US7716039B1 (en) * 2006-04-14 2010-05-11 At&T Intellectual Property Ii, L.P. Learning edit machines for robust multimodal understanding
US20080025617A1 (en) * 2006-07-25 2008-01-31 Battelle Memorial Institute Methods and apparatuses for cross-ontologial analytics
US20080077391A1 (en) * 2006-09-22 2008-03-27 Kabushiki Kaisha Toshiba Method, apparatus, and computer program product for machine translation
US20090070099A1 (en) * 2006-10-10 2009-03-12 Konstantin Anisimovich Method for translating documents from one language into another using a database of translations, a terminology dictionary, a translation dictionary, and a machine translation system
US20080114581A1 (en) * 2006-11-15 2008-05-15 Gil Meir Root cause analysis approach with candidate elimination using network virtualization
US20080133245A1 (en) * 2006-12-04 2008-06-05 Sehda, Inc. Methods for speech-to-speech translation
US20080154577A1 (en) * 2006-12-26 2008-06-26 Sehda,Inc. Chunk-based statistical machine translation system
US20090157380A1 (en) * 2007-12-18 2009-06-18 Electronics And Telecommunications Research Institute Method and apparatus for providing hybrid automatic translation
US20090265290A1 (en) * 2008-04-18 2009-10-22 Yahoo! Inc. Optimizing ranking functions using click data
US20090265230A1 (en) * 2008-04-18 2009-10-22 Yahoo! Inc. Ranking using word overlap and correlation features
US20100004920A1 (en) * 2008-07-03 2010-01-07 Google Inc. Optimizing parameters for machine translation
US20100004919A1 (en) * 2008-07-03 2010-01-07 Google Inc. Optimizing parameters for machine translation
US20100180244A1 (en) * 2009-01-15 2010-07-15 International Business Machines Corporation Method For Efficiently Checkpointing And Restarting Static Timing Analysis Of An Integrated Circuit Chip

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Franz Josef Och et a., "A Systematic Comparison of Various Statistical Alignment Models", 2003, Association for Computational Linguistics, pages 19-51 *
Hassan Sawaf et al., "On the Use of Grammar Based Language Models for Statistical Machine Translation", 1999 Nov. 5, pages 1-13 *

Cited By (97)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10198438B2 (en) 1999-09-17 2019-02-05 Sdl Inc. E-services translation utilizing machine translation and translation memory
US10216731B2 (en) 1999-09-17 2019-02-26 Sdl Inc. E-services translation utilizing machine translation and translation memory
US9954794B2 (en) 2001-01-18 2018-04-24 Sdl Inc. Globalization management system and method therefor
US10248650B2 (en) 2004-03-05 2019-04-02 Sdl Inc. In-context exact (ICE) matching
US10319252B2 (en) 2005-11-09 2019-06-11 Sdl Inc. Language capability assessment and training apparatus and techniques
US9798720B2 (en) 2008-10-24 2017-10-24 Ebay Inc. Hybrid machine translation
US20100228538A1 (en) * 2009-03-03 2010-09-09 Yamada John A Computational linguistic systems and methods
US9558183B2 (en) * 2009-09-04 2017-01-31 Synchronoss Technologies, Inc. System and method for the localization of statistical classifiers based on machine translation
US20120166183A1 (en) * 2009-09-04 2012-06-28 David Suendermann System and method for the localization of statistical classifiers based on machine translation
US9495127B2 (en) 2009-12-23 2016-11-15 Google Inc. Language model selection for speech-to-text conversion
US10157040B2 (en) 2009-12-23 2018-12-18 Google Llc Multi-modal input on an electronic device
US9031830B2 (en) 2009-12-23 2015-05-12 Google Inc. Multi-modal input on an electronic device
US9251791B2 (en) 2009-12-23 2016-02-02 Google Inc. Multi-modal input on an electronic device
US8751217B2 (en) 2009-12-23 2014-06-10 Google Inc. Multi-modal input on an electronic device
US9047870B2 (en) * 2009-12-23 2015-06-02 Google Inc. Context based language model selection
US10417646B2 (en) 2010-03-09 2019-09-17 Sdl Inc. Predicting the cost associated with translating textual content
US20110282643A1 (en) * 2010-05-11 2011-11-17 Xerox Corporation Statistical machine translation employing efficient parameter training
US8265923B2 (en) * 2010-05-11 2012-09-11 Xerox Corporation Statistical machine translation employing efficient parameter training
US9201871B2 (en) * 2010-06-11 2015-12-01 Microsoft Technology Licensing, Llc Joint optimization for machine translation system combination
US20110307244A1 (en) * 2010-06-11 2011-12-15 Microsoft Corporation Joint optimization for machine translation system combination
US8805669B2 (en) * 2010-07-13 2014-08-12 Dublin City University Method of and a system for translation
US20120016657A1 (en) * 2010-07-13 2012-01-19 Dublin City University Method of and a system for translation
US8818792B2 (en) * 2010-11-05 2014-08-26 Sk Planet Co., Ltd. Apparatus and method for constructing verbal phrase translation pattern using bilingual parallel corpus
US20130226556A1 (en) * 2010-11-05 2013-08-29 Sk Planet Co., Ltd. Machine translation device and machine translation method in which a syntax conversion model and a word translation model are combined
US10198437B2 (en) * 2010-11-05 2019-02-05 Sk Planet Co., Ltd. Machine translation device and machine translation method in which a syntax conversion model and a word translation model are combined
US20120150529A1 (en) * 2010-12-09 2012-06-14 Electronics And Telecommunication Research Institute Method and apparatus for generating translation knowledge server
US8352245B1 (en) 2010-12-30 2013-01-08 Google Inc. Adjusting language models
US8352246B1 (en) 2010-12-30 2013-01-08 Google Inc. Adjusting language models
US9542945B2 (en) 2010-12-30 2017-01-10 Google Inc. Adjusting language models based on topics identified using context
US9076445B1 (en) 2010-12-30 2015-07-07 Google Inc. Adjusting language models using context information
US8396709B2 (en) 2011-01-21 2013-03-12 Google Inc. Speech recognition using device docking context
US8296142B2 (en) 2011-01-21 2012-10-23 Google Inc. Speech recognition using dock context
US10061749B2 (en) 2011-01-29 2018-08-28 Sdl Netherlands B.V. Systems and methods for contextual vocabularies and customer segmentation
US10521492B2 (en) 2011-01-29 2019-12-31 Sdl Netherlands B.V. Systems and methods that utilize contextual vocabularies and customer segmentation to deliver web content
US20120209590A1 (en) * 2011-02-16 2012-08-16 International Business Machines Corporation Translated sentence quality estimation
US10140320B2 (en) 2011-02-28 2018-11-27 Sdl Inc. Systems, methods, and media for generating analytical data
US20140039879A1 (en) * 2011-04-27 2014-02-06 Vadim BERMAN Generic system for linguistic analysis and transformation
US8838434B1 (en) * 2011-07-29 2014-09-16 Nuance Communications, Inc. Bootstrap call router to other languages using selected N-best translations
US9984054B2 (en) 2011-08-24 2018-05-29 Sdl Inc. Web interface including the review and manipulation of a web document and utilizing permission based control
US20130054224A1 (en) * 2011-08-30 2013-02-28 Dublin City University Method and system for enhancing text alignment between a source language and a target language during statistical machine translation
US20150302001A1 (en) * 2012-02-16 2015-10-22 Continental Automotive Gmbh Method and device for phonetizing data sets containing text
US9158762B2 (en) 2012-02-16 2015-10-13 Flying Lizard Languages, Llc Deconstruction and construction of words of a polysynthetic language for translation purposes
US9436675B2 (en) * 2012-02-16 2016-09-06 Continental Automotive Gmbh Method and device for phonetizing data sets containing text
WO2013123384A3 (en) * 2012-02-16 2015-06-18 Standingdeer John C Deconstruction and construction of words of a polysynthetic language for translation purposes
US9213693B2 (en) * 2012-04-03 2015-12-15 Language Line Services, Inc. Machine language interpretation assistance for human language interpretation
US20130262079A1 (en) * 2012-04-03 2013-10-03 Lindsay D'Penha Machine language interpretation assistance for human language interpretation
US10402498B2 (en) 2012-05-25 2019-09-03 Sdl Inc. Method and system for automatic management of reputation of translators
US10261994B2 (en) 2012-05-25 2019-04-16 Sdl Inc. Method and system for automatic management of reputation of translators
US8583432B1 (en) 2012-07-18 2013-11-12 International Business Machines Corporation Dialect-specific acoustic language modeling and speech recognition
US20150287405A1 (en) * 2012-07-18 2015-10-08 International Business Machines Corporation Dialect-specific acoustic language modeling and speech recognition
US9966064B2 (en) * 2012-07-18 2018-05-08 International Business Machines Corporation Dialect-specific acoustic language modeling and speech recognition
US20140067361A1 (en) * 2012-08-28 2014-03-06 Xerox Corporation Lexical and phrasal feature domain adaptation in statistical machine translation
US9026425B2 (en) * 2012-08-28 2015-05-05 Xerox Corporation Lexical and phrasal feature domain adaptation in statistical machine translation
US10452740B2 (en) 2012-09-14 2019-10-22 Sdl Netherlands B.V. External content libraries
US9916306B2 (en) 2012-10-19 2018-03-13 Sdl Inc. Statistical linguistic analysis of source content
US9489351B2 (en) 2013-03-15 2016-11-08 The Dun & Bradstreet Corporation Enhancement of multi-lingual business indicia through curation and synthesis of transliteration, translation and graphemic insight
CN105210057A (en) * 2013-03-15 2015-12-30 邓白氏公司 Multi-lingual business indicia curation and transliteration synthesis
WO2014144716A1 (en) * 2013-03-15 2014-09-18 The Dun & Bradstreet Corporation Multi-lingual business indicia curation and transliteration synthesis
RU2644071C2 (en) * 2013-03-15 2018-02-07 Дзе Дан Энд Брэдстрит Корпорейшн Curation of multilingual commercial indicators and synthesis of transliteration
US20150066503A1 (en) * 2013-08-28 2015-03-05 Verint Systems Ltd. System and Method of Automated Language Model Adaptation
US9633650B2 (en) * 2013-08-28 2017-04-25 Verint Systems Ltd. System and method of automated model adaptation
US9508346B2 (en) * 2013-08-28 2016-11-29 Verint Systems Ltd. System and method of automated language model adaptation
US9990920B2 (en) 2013-08-28 2018-06-05 Verint Systems Ltd. System and method of automated language model adaptation
US20150066502A1 (en) * 2013-08-28 2015-03-05 Verint Systems Ltd. System and Method of Automated Model Adaptation
US9779724B2 (en) 2013-11-04 2017-10-03 Google Inc. Selecting alternates in speech recognition
US10140978B2 (en) 2013-11-04 2018-11-27 Google Llc Selecting alternates in speech recognition
US20150178274A1 (en) * 2013-12-25 2015-06-25 Kabushiki Kaisha Toshiba Speech translation apparatus and speech translation method
US9653071B2 (en) * 2014-02-08 2017-05-16 Honda Motor Co., Ltd. Method and system for the correction-centric detection of critical speech recognition errors in spoken short messages
US20150228272A1 (en) * 2014-02-08 2015-08-13 Honda Motor Co., Ltd. Method and system for the correction-centric detection of critical speech recognition errors in spoken short messages
US9842592B2 (en) 2014-02-12 2017-12-12 Google Inc. Language models using non-linguistic context
US9805031B2 (en) 2014-02-28 2017-10-31 Ebay Inc. Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US9530161B2 (en) 2014-02-28 2016-12-27 Ebay Inc. Automatic extraction of multilingual dictionary items from non-parallel, multilingual, semi-structured data
US9569526B2 (en) 2014-02-28 2017-02-14 Ebay Inc. Automatic machine translation using user feedback
US9940658B2 (en) 2014-02-28 2018-04-10 Paypal, Inc. Cross border transaction machine translation
US9881006B2 (en) 2014-02-28 2018-01-30 Paypal, Inc. Methods for automatic generation of parallel corpora
US9412365B2 (en) 2014-03-24 2016-08-09 Google Inc. Enhanced maximum entropy models
US10013417B2 (en) 2014-06-11 2018-07-03 Facebook, Inc. Classifying languages for objects and entities
US10452786B2 (en) 2014-12-29 2019-10-22 Paypal, Inc. Use of statistical flow data for machine translations between different languages
EP3062238A1 (en) * 2015-02-27 2016-08-31 Samsung Electronics Co., Ltd. Summarization by sentence extraction and translation of summaries containing named entities
US10134394B2 (en) 2015-03-20 2018-11-20 Google Llc Speech recognition using log-linear model
WO2017033063A3 (en) * 2015-08-25 2017-04-27 Alibaba Group Holding Limited Statistics-based machine translation method, apparatus and electronic device
US10255275B2 (en) 2015-08-25 2019-04-09 Alibaba Group Holding Limited Method and system for generation of candidate translations
US10268685B2 (en) * 2015-08-25 2019-04-23 Alibaba Group Holding Limited Statistics-based machine translation method, apparatus and electronic device
US20170060854A1 (en) * 2015-08-25 2017-03-02 Alibaba Group Holding Limited Statistics-based machine translation method, apparatus and electronic device
US10346537B2 (en) 2015-09-22 2019-07-09 Facebook, Inc. Universal translation
US10409913B2 (en) 2015-10-01 2019-09-10 Conduent Business Services, Llc Methods and systems to train classification models to classify conversations
US10089299B2 (en) 2015-12-17 2018-10-02 Facebook, Inc. Multi-media context language processing
US9747283B2 (en) * 2015-12-28 2017-08-29 Facebook, Inc. Predicting future translations
US20170185588A1 (en) * 2015-12-28 2017-06-29 Facebook, Inc. Predicting future translations
US10289681B2 (en) * 2015-12-28 2019-05-14 Facebook, Inc. Predicting future translations
US9978367B2 (en) 2016-03-16 2018-05-22 Google Llc Determining dialog states for language models
US10185763B2 (en) * 2016-11-30 2019-01-22 Facebook, Inc. Syntactic models for parsing search queries on online social networks
US10311860B2 (en) 2017-02-14 2019-06-04 Google Llc Language model biasing system
US20180373704A1 (en) * 2017-06-21 2018-12-27 Samsung Electronics Co., Ltd. Method and apparatus for machine translation using neural network and method of training the apparatus
US10474758B2 (en) * 2017-06-21 2019-11-12 Samsung Electronics Co., Ltd. Method and apparatus for machine translation using neural network and method of training the apparatus
US10540450B2 (en) 2017-09-05 2020-01-21 Facebook, Inc. Predicting future translations
US10380249B2 (en) 2017-10-02 2019-08-13 Facebook, Inc. Predicting future trending topics

Also Published As

Publication number Publication date
WO2010046782A2 (en) 2010-04-29
US9798720B2 (en) 2017-10-24
WO2010046782A3 (en) 2010-06-17

Similar Documents

Publication Publication Date Title
Barzilay et al. Extracting paraphrases from a parallel corpus
Vilar et al. Error Analysis of Statistical Machine Translation Output.
Gao et al. Chinese word segmentation and named entity recognition: A pragmatic approach
US6952665B1 (en) Translating apparatus and method, and recording medium used therewith
Marino et al. N-gram-based machine translation
US6760695B1 (en) Automated natural language processing
US6266642B1 (en) Method and portable apparatus for performing spoken language translation
US8719006B2 (en) Combined statistical and rule-based part-of-speech tagging for text-to-speech synthesis
JP3220560B2 (en) Machine translation apparatus
US6442524B1 (en) Analyzing inflectional morphology in a spoken language translation system
US4864503A (en) Method of using a created international language as an intermediate pathway in translation between two national languages
US6278967B1 (en) Automated system for generating natural language translations that are domain-specific, grammar rule-based, and/or based on part-of-speech analysis
US7319949B2 (en) Unilingual translator
US6278968B1 (en) Method and apparatus for adaptive speech recognition hypothesis construction and selection in a spoken language translation system
US20050216253A1 (en) System and method for reverse transliteration using statistical alignment
US6233546B1 (en) Method and system for machine translation using epistemic moments and stored dictionary entries
US20170177563A1 (en) Methods and systems for automated text correction
US6223150B1 (en) Method and apparatus for parsing in a spoken language translation system
US6356865B1 (en) Method and apparatus for performing spoken language translation
EP1471439A1 (en) Syntax analysis method and apparatus
US6282507B1 (en) Method and apparatus for interactive source language expression recognition and alternative hypothesis presentation and selection
US6243669B1 (en) Method and apparatus for providing syntactic analysis and data structure for translation knowledge in example-based language translation
JP4237001B2 (en) System and method for automatically detecting collocation errors in documents
US6374224B1 (en) Method and apparatus for style control in natural language generation
US8386234B2 (en) Method for generating a text sentence in a target language and text sentence generating apparatus

Legal Events

Date Code Title Description
AS Assignment

Owner name: APPTEK, VIRGINIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SAWAF, HASSAN;SHIHADAH, MOHAMMAD;YAGHI, MUDAR;REEL/FRAME:024105/0431

Effective date: 20091130

AS Assignment

Owner name: SCIENCE APPLICATIONS INTERNATIONAL CORPORATION, CA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:APPLICATIONS TECHNOLOGY, INC.;REEL/FRAME:025414/0565

Effective date: 20101110

AS Assignment

Owner name: LEIDOS, INC., VIRGINIA

Free format text: CHANGE OF NAME;ASSIGNOR:SCIENCE APPLICATIONS INTERNATIONAL CORPORATION;REEL/FRAME:031534/0765

Effective date: 20130927

AS Assignment

Owner name: APPLICATIONS TECHNOLOGY (APPTEK), LLC, VIRGINIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:LEIDOS, INC.;REEL/FRAME:031541/0926

Effective date: 20131029

AS Assignment

Owner name: APPLICATIONS TECHNOLOGY (APPTEK), LLC, VIRGINIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SAWAF, HASSAN;SHIHADAH, MOHAMMAD;YAGHI, MUDAR;REEL/FRAME:033068/0901

Effective date: 20091129

AS Assignment

Owner name: EBAY INC., CALIFORNIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:APPLICATIONS TECHNOLOGY (APPTEK), LLC;REEL/FRAME:033293/0752

Effective date: 20140612

Owner name: APPLICATIONS TECHNOLOGY (APPTEK), LLC, VIRGINIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:SAWAF, HASSAN;SHIHADAH, MOHAMMAD;YAGHI, MUDAR;REEL/FRAME:033293/0721

Effective date: 20091130

FEPP Fee payment procedure

Free format text: ENTITY STATUS SET TO UNDISCOUNTED (ORIGINAL EVENT CODE: BIG.)

STCF Information on status: patent grant

Free format text: PATENTED CASE